I've been reading about mathematical theories of "evidence" - how one should draw conclusions based on different types of (perhaps conflicting) evidence. Dempster-Schafer seems to be the big name, although there are a few offshoots from this that deal with different areas where DS doesn't seem to produce intuitively correct outputs. I find it interesting that "science" is so fundamentally based on the concept of experimental support for theories, but that there isn't a good single theory for describing how evidence supports conclusions.
If you type Dempster Schafer into Google, the first result is a DSTO report, which outlines the concepts nicely, and has a good comparison with Bayes' probabilities.
Although I'm a Wikipedia fan, its article in this area isn't so good, I don't think. The best thing was that it pointed me (in the last reference) to a very clearly written Masters thesis which sums it all up nicely.
Oh, and by the way, it is actually work-related! I feel lucky to be able to read about such interesting things in my work, but a bit embarrassed that I wasn't already aware of such things.
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment